Detecting cyber attacks with packet loss resilience for power systems

Phasor measurement units (PMUs) are critical for modern power systems to monitor the operation status in near real-time. Spoof attacks present increasing challenges to the trustworthy of the PMU data. Machine learning models can be used to detect spoofed signals in a timely manner. However, the effectiveness of these methods can be degraded by the unpredictable packet loss during the transmission of the PMU data. The negative impact of the packet loss on machine learning based spoof detection models have not been well studied. In this paper, the exact granular effects of packet loss on spoof detection models is quantified and examined. Several mitigation strategies are considered and recommended when dealing with packet loss in the data stream. Finally, a Dynamic Window Size Algorithm (DWSA) is developed to minimize the amount of packet loss the models are passed for each data point. The positive and negative effects of using DWSA to mitigate the effects of packet loss are also explored.